TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Show HN: Bayesian AB-test analysis only needs 4 numbers

2 pointsby KaseKunabout 2 years ago
Hey (:)<p>You&#x27;d be surprised how many of the (quite talented) data analysts I&#x27;ve worked with don&#x27;t truly understand the beautiful simplicity of Bayesian AB testing.<p>When AB testing, most metrics we care about are binomial (i.e., did or didn&#x27;t happen -&gt; 1 or 0). Think &quot;User clicked login&quot; or &quot;User upgraded to Premium&quot;.<p>This means if you want to build an API for statistically analyzing these experiments all you need as inputs are 4 numbers (per metric): - Control: Total users, and users who &quot;did&quot; - Variant: Total users and users who &quot;did&quot;<p>Then you can reconstruct the &quot;observation array&quot; on the other side of the API, statistically compare your control to your variant, and send the results back!<p>Right now, if you have a way to pull those 4 numbers from Amplitude, or your database, you can run a Bayesian analysis using the analyzer i built.<p>Have fun!<p>rate limits do apply, product translation sold separately

no comments

no comments